Re: [OJB-developers] ODMG transaction question

[Thomas M. <tho...@ho...>]

Hi Matt,

> Matt Goodall wrote:
> 
> Hi,
> 
> I have found something which I can only describe as strange, perhaps
> there is a legitimate explanation or perhaps it's a bug.
> 
> First of all, some example code ...
> 
> Database schema:
> 
>     create table artist
>     (
>         id int not null,
>         name varchar2(100),
>         primary key (id),
>         unique (name)
>     )
> Code:
> 
>     Transaction tx = odmg.newTransaction();
>     tx.begin();
> 
>     Artist artist = new Artist();
>     artist.name = "xyz";
>     tx.lock(artist, Transaction.WRITE);
>     tx.checkpoint();
> 
>     OQLQuery query = odmg.newOQLQuery();
>     query.create("select allArtists from ojbtest.Artist");
>     Collection list = (Collection) query.execute();
>     Iterator it = list.iterator();
>     while(it.hasNext())
>     {
>         artist = (Artist) it.next();
>         System.out.println(artist);
>     }
> 
>     tx.commit();
> 
> The 1st time the code is run the checkpoint commits the artist to the
> database, as expected, and the "select allArtists" query returns the
> expected rows.
> 
> The 2nd time through, the unique(name) constraint means that the
> artist cannot be added but the checkpoint succeeds (!!!), the "select
> allArtists" returns the same objects that were seen the 1st time and,
> finally, an exception is thrown in tx.commit().
> 
> As a test I wrote almost the equivalent using the lower level OJB
> interface and everything worked as expected i.e. an exception was
> thrown as soon as I tried to add the same artist.
>
> Is this the correct behaviour or is it a bug?

This is definitively a bug in .checkpoint() !
I assume that exceptions thrown from the lower level PersistenceBroker
API are ignored or not handled correctly during .checkpoint().

> 
> I have to say, I'm not sure I'd ever use transactions like this, it
> all came about through some test code which *always* aborted the
> transaction. I didn't have the checkpoint() then either. The intention
> was to leave the database intact after running the test which addded
> data and then retrieved it again - I was just playing with the API
> basically.
> 

But even if it's no realistic scenario: OJB must handle it correctly!!!

Thanks for you detailed report, it will help me to fix this thing
quickly!

--Thomas

> Cheers, Matt
> ---
> Matt Goodall
> Analyst Programmer, Strategic Systems Solutions
> e: mat...@ss..., t: +44 113 3892645
>